Not necessarily. The task of assembling them into an attack is still a significant investment. And once that is done you face the question of whether it is more profitable to mine or attack. Empirically it seems the incentives are usually to mine, with a few (so far) outlier exceptions (low-usage coins with large rapid drops in hash rate).

You mentioned Google or NSA, but both Google and NSA have those computers for a reason, so presumably they are already doing something. If you want to take those computers and use them to attack a coin, that has a significant opportunity cost. And even at that, Google still isn't that big. XMR is reasonably close (1-2 orders of magnitude) to exceeding Google's rumored entire 1m computer capacity (meaning you would have to shut down Google to pull off the attack -- good luck with that plan), and XMR is still a tiny coin.

I think you underestimate the task of building a large attack. By contrast, any subset of these computers can just go ahead and mine instead